Agent Graphs
For agentic workflows, a flat observation tree doesn't always convey how an agent moved between steps. The Graph view renders a trace as a directed graph — nodes are agent steps, edges are the execution flow — so you can see the path the agent actually took.
ℹ
The Graph view appears as a tab on a trace when graph data is present. It's designed for LangGraph-based agents, which emit node/step metadata the platform reads automatically.
What you see
- Nodes — each agent step (from the agent's node metadata), including special start and end markers.
- Edges — the order and branching of execution between steps.
- Selection — click a node to highlight it and open the underlying observation for full input/output detail.
Getting graph data
Graph rendering relies on agent-step metadata on your observations. When you instrument a LangGraph agent (see LangChain & LangGraph), each node/step is tagged automatically and the Graph tab lights up — no extra configuration.
When to use it
- Understand control flow — see loops, branches, and retries an agent took.
- Find where it derailed — spot the node where an agent went off course, then click in to inspect that step.
Next steps
- Traces & Observations — the detail behind each node.
- LangChain & LangGraph — instrument agents to populate the graph.